linear discriminant analysis

Community Extensions
1

Hi guys,

can you please help me with my linear disriminant analysis?

I have made some prediction (churn 0/1) via gradient boosted trees and now want to find out, which of the features are discriminant. 

I have about 130 features that are numerical, date-time, nominal -> is that ok for LDA? or are there too many features or should I only use numerical ones?

Thanks in advance

Greetings Daniel 

2

Hi Daniel, 

You have to use continuous independent variables and a categorical dependent variable for LDA. You can use the One To Many node to create dummy variables, however it will significantly increase the number of features. 

The number of features to use should be guided by the characteristics of the dataset you're working with. You can also check KNIME Feature Elimination meta nodes. 

Best,

Anna